DOE for PRA Testbed

Report
Design of Experiments and the Probability of
Raid Annihilation (PRA) Testbed
Presenter: Richard Lawrence
860 Greenbrier Circle
Suite 305
Chesapeake, VA 23320
www.avwtech.com
Phone:
757-361-9581
Fax:
757-361-9585
1
AVW Technologies, Inc
Design of Experiments and PRA Testbed
Introduction
Design of Experiments (DOE) offers the opportunity for efficiency in test
execution and to gain insight to the operations of complex systems. The current
AAW SSD PRA metric is not DOE friendly in that the outputs of the Testbed do not
lend themselves to straightforward statistical analysis. This is an overview of the
challenges to executing a serious DOE process on the PRA Testbed and proposed
methods to solve these challenges.
Presentation Outline
- Usable definition of DOE
- Scoring
- Background
- Non-Determinism
- Challenges
- Basic Steps
- Levels of Factors
- Conclusion
2
AVW Technologies, Inc
Principles of DOE
• Create a statistical model of a system based on identified
factors and measured outputs.
• Purposefully vary input (factors) and correlate with outputs.
X1
Y1
X2
Inputs
X3
PROCESS
Y2
Outputs
X4
3
AVW Technologies, Inc
Principles of DOE (2)
• Run matrix is developed from several techniques to
establish a ‘sample space’.
• Different from ‘One Factor at a Time’ testing because
variables are changed several at a time and effects are
separated in the analysis phase.
• Through analysis of outputs, identifies and quantifies effects
of various factors
Run
X1
X2
X3
X4
1
+
0
0
5
2
-
+
+
2
3
-
0
+
2
Y1
Y2
Yave
σY
n
4
AVW Technologies, Inc
Background
• Current PRA methodology is:
•Operationally relevant
•Accepted and Established
•Oriented solely to calculate a single value (PRA)
•Ideal candidate for in-depth DOE
T1R1 - sea-skimming, subsonic RF threat
T2 - sea-skimming, subsonic Imaging IR threat
Scenario
“Clean”
and
“Dirty”
Signatures
AVW Technologies, Inc
T5 - high diver, supersonic RF ARM threat
T7 - sea-skimming, maneuvering supersonic
Advanced RF Threat
Littoral Scenario
5
Background and Challenges
• Run Reduction Strategy for LPD 17 based on rudimentary
application of DOE.
• Historical analysis was attempted for LPD 17 data by AVW
and DOT&E.
• No surprising insights resulted
• Lessons learned:
•Categorized variables like radial are difficult to analyze,
since they are not continuous.
•Binary outcomes are even more difficult to analyze
because there is no conventional way to calculate
variance & other statistical parameters.
• Necessitates a more in-depth approach to find confounding
6
factors and their interrelationships.
AVW Technologies, Inc
Additional Challenges
• Difficult to identify ‘specific factors’ in particular scenarios
•Requires runs to investigate
• Each scenario (radial, TOD, etc.) has a confluence of factors
• Example--different radials vary the following:
• RF propagation for ship sensors (duct strength, height)
• RF clutter for ship radars
(land, wave direction)
• Ship radar blockage
• RF propagation for threat seeker
• Ship RCS/Decoy effectiveness
• IR scene for RAM
• Wind
• Threat spacing in bearing and distance
7
AVW Technologies, Inc
Example: Environment Categorization
• Empirical way to quantify input factors
• Each radial would include a parameter for RF prop, clutter, IR scene,
RCS, threat separation
CNR vs. Range for 8 Radials from 20 to 335 Degrees True
• Categorize each parameter:
100
20T
65T
110T
155T
220T
245T
290T
335T
80
• +1 favorable conditions
60
• 0 neutral conditions
• -1 adverse conditions
CNR (dB)
40
20
0
-20
-40
-60
-80
0
10
20
30
40
50
60
70
range (km)
Ultimate goal is to
eliminate test cases.
8
AVW Technologies, Inc
Scoring
• Analytical way to quantify outcomes
• Miss distance
• Aimpoint errors
• Scoring related to ship (vulnerability)
9
AVW Technologies, Inc
Non-Determinism
• Demonstrated differing outcomes given
identical scenarios during DT5.
• Attributable to the way tactical software
operates.
• Is there a minimum number of trials required to
give a statistically significant outcome?
•Would
require a large number of runs for
each scenario
OR
• Treat each scenario as we do in real ships—
any given event can go many different ways
10
AVW Technologies, Inc
Basic Steps
• Identify factors
• Establish types and levels for each factor
• Assign factor levels to each scenario
• Identify, execute and analyze screening runs
• Develop formalized run matrix
• Execute runs
• Analyze results
• Refine formalized run matrix
based on identified relevant
factors
11
AVW Technologies, Inc
Bottom Lines
• DOE is a natural complement to ongoing V&V process
• Gain value from Testbed runs during DTs (maximizing
resource investment)
• Analytical insight into Combat Systems performance and
factors influencing engagement outcomes
• Defendable approach to Testbed runs—to complement
COTF’s methodology
Test
Analyze
12
AVW Technologies, Inc
Design of Experiments and the Probability of
Raid Annihilation (PRA) Testbed
860 Greenbrier Circle
Suite 305
Chesapeake, VA 23320
www.avwtech.com
Phone:
757-361-9581
Fax:
757-361-9585
13
AVW Technologies, Inc

similar documents